Re: [Pharo-dev] GTDebugger variables table

[stepharong]

On Thu, 02 Feb 2017 17:16:49 +0100, John Brant  
 wrote:



On 02/02/2017 04:22 AM, Denis Kudriashov wrote:


Now I think I realized main reason of my confusion. Temps and receiver
vars are not just in single table but they are also sorted by name all
together.


I'm not a fan of this either. While I can filter by the type of variable  
to limit the list, the next time I step the debugger, the whole list  
resets. The list filter and selection should be kept across steps in the  
debugger (if possible).


Some might argue that you should have fewer variables so the list would  
be easier to use in the debugger. However, if you are using the  
debugger, likely you are still in the "make it work" phase and haven't  
performed the factoring from the "make it right" phase.


+ 1 and under stress


I think my preference would be to have several tabs for the variables.  
In addition to the one tab that we have now that shows all variables, I  
can think of tabs for locals, inst vars, interesting variables, watched  
variables/expressions, and stack variables. Locals would show just the  
method/block arguments and any temps defined in the method. Inst vars  
would show the object's inst vars (and maybe class vars, however these  
would only appear after the inst vars).


Interesting variables would show locals and inst vars used by the  
method. The locals would be limited to the ones that are still active at  
the current location in the method. For example, if you are in a block,  
it would only show variables used in the block. Also, if you are  
before(/after) the first(/final) use of the variable, it wouldn't show  
in the interesting list. Interesting variables should also do some  
analysis to see what accessor methods are used and show their  
corresponding variables.


I like that



Watched variables/expressions would be user controlled. The user could  
add/remove variables or expressions. These variables/expressions would  
remain across different methods. If a variable didn't exist in the other  
method, "Invalid" could be displayed.


Finally, stack variables would display the whole stack and not just the  
top item. I like the ability to see the stack top, but it really doesn't  
work if you want to see the first argument of a two argument message  
send. For example, if you debug "Array with: OrderedCollection new with:  
Set new", stepping over the "OrderedCollection new" immediately pushes  
the "Set" class on the stack so you can't see the "OrderedCollection  
new" object.


I like that now when we have too many tabs it often happens that
only two are useful.



BTW, the current variable list sometimes shows 'error obtaining field  
value' for temporaries when stepping through a method. I'm not sure why  
it occurs, but it should always be able to display the temp variables.



John Brant




--
Using Opera's mail client: http://www.opera.com/mail/

Re: [Pharo-dev] GTDebugger variables table

[Ben Coman]

On Fri, Feb 3, 2017 at 10:21 AM, Ben Coman  wrote:

> On Fri, Feb 3, 2017 at 12:28 AM, Denis Kudriashov 
> wrote:
> >
> > 2017-02-02 17:16 GMT+01:00 John Brant :
> >>
> >> I think my preference would be to have several tabs for the variables.
> In
> >> addition to the one tab that we have now that shows all variables, I can
> >> think of tabs for locals, inst vars, interesting variables, watched
> >> variables/expressions, and stack variables. Locals would show just the
> >> method/block arguments and any temps defined in the method.
> >> Inst vars would
> >> show the object's inst vars (and maybe class vars, however these would
> only
> >> appear after the inst vars).
> >>
> >> Interesting variables would show locals and inst vars used by the
> method.
> >> The locals would be limited to the ones that are still active at the
> current
> >> location in the method. For example, if you are in a block, it would
> only
> >> show variables used in the block.
>
> >> Also, if you are before(/after) the first(/final) use of the variable,
> >> it wouldn't show in the interesting list.
>
> I'm not sure I'd like variables slipping in and out and rearranging
> the middle of the list I'm looking at.  Perhaps scope could be
> indicated another way like greying out variables.  It would be okay
> for block variables to be added at the bottom.
>
> >> Interesting variables should also do some analysis to see what accessor
> >> methods are used and show their corresponding variables.
>
> This is a nice idea.  It need not only be for interesting variables.
> Ideally it would be a tab in the next pane to the right of the [Meta]
> tab, but I guess its difficult since that pane is concerned with the
> object, and linking back to the variable holding could be difficult.
>
> >>
> >> Watched variables/expressions would be user controlled. The user could
> >> add/remove variables or expressions. These variables/expressions would
> >> remain across different methods. If a variable didn't exist in the other
> >> method, "Invalid" could be displayed.
>
> I really miss this from my TurboPascal days.
> The difficulty of course is avoiding side effects.
>
> >>
> >> Finally, stack variables would display the whole stack and not just the
> >> top item. I like the ability to see the stack top, but it really doesn't
> >> work if you want to see the first argument of a two argument message
> send.
> >> For example, if you debug "Array with: OrderedCollection new with: Set
> new",
> >> stepping over the "OrderedCollection new" immediately pushes the "Set"
> class
> >> on the stack so you can't see the "OrderedCollection new" object.
>
> Perhaps the stack could be another tab next to [Variables] ?
> Or maybe it [Stack] would be better as a tab of thisContext's next pane.
>
>
Actually you can try...

Context>>stackValueMap
| stackValues stackDepth|
stackValues := OrderedCollection new.
stackDepth := self basicSize min: 21.
(stackDepth to: 1 by: -1) do: [ :index |
|key|
key := (index = stackDepth) ifTrue: ['Top'] ifFalse: [(stackDepth - index +
1) asString].
stackValues add: (key -> (self basicAt: index))
].
^stackValues

Context>>gtInspectorStackValuesIn: composite

| stackValues |
stackValues := self stackValues.
^ (composite table)
title: 'Stack';
display: [ self stackValueMap ];
column: 'Depth' evaluated: [ :assoc | assoc key ] width: 50;
column: 'Item' evaluated: [ :assoc | GTObjectPrinter new
asTruncatedTextFrom: assoc value ]


I'm not sure whether the #stackValueMap is an appropriate name. Perhaps it
should be #methodStackValueMap or something else?  Similar for the tab
name.

And I don't know what the magic number 21 is, as I queried here...
http://forum.world.st/magic-numbers-in-gtInspectorVariableValuePairs-tp4932831.html

cheers -ben

Re: [Pharo-dev] GTDebugger variables table

[Ben Coman]

On Fri, Feb 3, 2017 at 12:28 AM, Denis Kudriashov  wrote:
>
> 2017-02-02 17:16 GMT+01:00 John Brant :
>>
>> I think my preference would be to have several tabs for the variables. In
>> addition to the one tab that we have now that shows all variables, I can
>> think of tabs for locals, inst vars, interesting variables, watched
>> variables/expressions, and stack variables. Locals would show just the
>> method/block arguments and any temps defined in the method.
>> Inst vars would
>> show the object's inst vars (and maybe class vars, however these would only
>> appear after the inst vars).
>>
>> Interesting variables would show locals and inst vars used by the method.
>> The locals would be limited to the ones that are still active at the current
>> location in the method. For example, if you are in a block, it would only
>> show variables used in the block.

>> Also, if you are before(/after) the first(/final) use of the variable,
>> it wouldn't show in the interesting list.

I'm not sure I'd like variables slipping in and out and rearranging
the middle of the list I'm looking at.  Perhaps scope could be
indicated another way like greying out variables.  It would be okay
for block variables to be added at the bottom.

>> Interesting variables should also do some analysis to see what accessor
>> methods are used and show their corresponding variables.

This is a nice idea.  It need not only be for interesting variables.
Ideally it would be a tab in the next pane to the right of the [Meta]
tab, but I guess its difficult since that pane is concerned with the
object, and linking back to the variable holding could be difficult.

>>
>> Watched variables/expressions would be user controlled. The user could
>> add/remove variables or expressions. These variables/expressions would
>> remain across different methods. If a variable didn't exist in the other
>> method, "Invalid" could be displayed.

I really miss this from my TurboPascal days.
The difficulty of course is avoiding side effects.

>>
>> Finally, stack variables would display the whole stack and not just the
>> top item. I like the ability to see the stack top, but it really doesn't
>> work if you want to see the first argument of a two argument message send.
>> For example, if you debug "Array with: OrderedCollection new with: Set new",
>> stepping over the "OrderedCollection new" immediately pushes the "Set" class
>> on the stack so you can't see the "OrderedCollection new" object.

Perhaps the stack could be another tab next to [Variables] ?
Or maybe it [Stack] would be better as a tab of thisContext's next pane.

>>
>>
>> BTW, the current variable list sometimes shows 'error obtaining field
>> value' for temporaries when stepping through a method. I'm not sure why it
>> occurs, but it should always be able to display the temp variables.
>
>
> I want everything you suggest :)

So we should try in Pharo 7.
cheers -ben

Re: [Pharo-dev] GTDebugger variables table

[Denis Kudriashov]

2017-02-02 17:16 GMT+01:00 John Brant :

> I think my preference would be to have several tabs for the variables. In
> addition to the one tab that we have now that shows all variables, I can
> think of tabs for locals, inst vars, interesting variables, watched
> variables/expressions, and stack variables. Locals would show just the
> method/block arguments and any temps defined in the method. Inst vars would
> show the object's inst vars (and maybe class vars, however these would only
> appear after the inst vars).
>
> Interesting variables would show locals and inst vars used by the method.
> The locals would be limited to the ones that are still active at the
> current location in the method. For example, if you are in a block, it
> would only show variables used in the block. Also, if you are
> before(/after) the first(/final) use of the variable, it wouldn't show in
> the interesting list. Interesting variables should also do some analysis to
> see what accessor methods are used and show their corresponding variables.
>
> Watched variables/expressions would be user controlled. The user could
> add/remove variables or expressions. These variables/expressions would
> remain across different methods. If a variable didn't exist in the other
> method, "Invalid" could be displayed.
>
> Finally, stack variables would display the whole stack and not just the
> top item. I like the ability to see the stack top, but it really doesn't
> work if you want to see the first argument of a two argument message send.
> For example, if you debug "Array with: OrderedCollection new with: Set
> new", stepping over the "OrderedCollection new" immediately pushes the
> "Set" class on the stack so you can't see the "OrderedCollection new"
> object.
>
>
> BTW, the current variable list sometimes shows 'error obtaining field
> value' for temporaries when stepping through a method. I'm not sure why it
> occurs, but it should always be able to display the temp variables.
>

I want everything you suggest :)

Re: [Pharo-dev] GTDebugger variables table

[John Brant]

On 02/02/2017 04:22 AM, Denis Kudriashov wrote:


Now I think I realized main reason of my confusion. Temps and receiver
vars are not just in single table but they are also sorted by name all
together.


I'm not a fan of this either. While I can filter by the type of variable 
to limit the list, the next time I step the debugger, the whole list 
resets. The list filter and selection should be kept across steps in the 
debugger (if possible).


Some might argue that you should have fewer variables so the list would 
be easier to use in the debugger. However, if you are using the 
debugger, likely you are still in the "make it work" phase and haven't 
performed the factoring from the "make it right" phase.




I think we can improve this part of debugger. My idea that variable
table should show only used temp and variables. And "self" should be
selected by default.
With this main table will show only important information. And on the
right pane we will see receiver state like in the old debuggers.

What you think?


I think my preference would be to have several tabs for the variables. 
In addition to the one tab that we have now that shows all variables, I 
can think of tabs for locals, inst vars, interesting variables, watched 
variables/expressions, and stack variables. Locals would show just the 
method/block arguments and any temps defined in the method. Inst vars 
would show the object's inst vars (and maybe class vars, however these 
would only appear after the inst vars).


Interesting variables would show locals and inst vars used by the 
method. The locals would be limited to the ones that are still active at 
the current location in the method. For example, if you are in a block, 
it would only show variables used in the block. Also, if you are 
before(/after) the first(/final) use of the variable, it wouldn't show 
in the interesting list. Interesting variables should also do some 
analysis to see what accessor methods are used and show their 
corresponding variables.


Watched variables/expressions would be user controlled. The user could 
add/remove variables or expressions. These variables/expressions would 
remain across different methods. If a variable didn't exist in the other 
method, "Invalid" could be displayed.


Finally, stack variables would display the whole stack and not just the 
top item. I like the ability to see the stack top, but it really doesn't 
work if you want to see the first argument of a two argument message 
send. For example, if you debug "Array with: OrderedCollection new with: 
Set new", stepping over the "OrderedCollection new" immediately pushes 
the "Set" class on the stack so you can't see the "OrderedCollection 
new" object.



BTW, the current variable list sometimes shows 'error obtaining field 
value' for temporaries when stepping through a method. I'm not sure why 
it occurs, but it should always be able to display the temp variables.



John Brant

Re: [Pharo-dev] GTDebugger variables table

[Ben Coman]

> On 02/02/2017 11:22, Denis Kudriashov wrote:
>
> Hi.
>
> Finally I force myself to report my bad feeling of merged variable table in
> GTDebugger.
> By "merged" I mean that debugger join temps and receiver state in one table.
> Sometimes I really not like it because it is difficult to find concrete
> variable.
>
> Now I think I realized main reason of my confusion. Temps and receiver vars
> are not just in single table but they are also sorted by name all together.
>
> For example try debug #expandBy: method:
>
> (1@2 corner: 3@4) expandBy: 10
>
> You will see rows: self, corner, delta, origin, thisContext, stackTop
> Maybe in this example it is not really bad. But it shows problem.
> And imaging that there are much more inst vars in receiver object like in
> morph or Spec. It is really difficult to find desired temp.
> Actually it is also difficult to find inst var which are really used in
> selected method. And usually methods use only few variables and few temps.
> And the rest variables in table are just waste.
>
> I think we can improve this part of debugger. My idea that variable table
> should show only used temp and variables. And "self" should be selected by
> default.
> With this main table will show only important information. And on the right
> pane we will see receiver state like in the old debuggers.
>
> What you think?

I think thats a really innovative approach.  I've haven't been too
bothered by this, but I imagine it would feel like an improvement.
I'd like to try it.

On Thu, Feb 2, 2017 at 7:00 PM, Nicolas Anquetil
 wrote:
> for what it is worth, Eclipse behaves more or less as described by Denis:
>
> you see only local variables (self being one) and if a variable contains an
> object (such as self), it expands in a three to show the attributes of this
> object.

in a tree?

>
> I sometimes find it a pain to have to expand "self" to see the attributes,

but in this case it would be pre-selected and already showing in the second pane
cheers -ben

> but it is true that when there are many attributes/local variables, one gets
> easily annoyed too.